A wave that is not quite what it looks like
In 2025, artificial intelligence was cited as a factor in approximately 55,000 layoffs in the United States alone. For 2026, the five largest hyperscalers will spend close to $700 billion on AI-related capital expenditures, almost twice the 2025 figure. And in the same period, MIT NANDA research found that 95% of enterprise AI pilots produced no measurable impact on P&L.
These three numbers tell three stories. The standard narrative bundles them as cause and effect. AI is changing work, companies are investing aggressively, and layoffs are the inevitable consequence. The reality is more uncomfortable.
Three weeks ago, sitting in a boardroom with a view of a half-occupied operations floor, I watched a CFO ask the only question that mattered. Hiring had been frozen. Discretionary spend was tight. Budget compliance was within 1.8%. HR reported headcount stability. Operations argued productivity was up. And yet, margins had dropped by 4.2 points over two quarters.
“If everything is under control,” the CFO said quietly, “where is the money going?”
No one answered. The data was there. It was telling a story none of the planning instruments in the room could read. The model itself was broken. Across thirty engagements like that one, I have learned that most of the AI-justified layoffs you have been reading about come from organizations computing an old equation under pressure to compute a new one. Few of them are strategic capability decisions.
For decades, every workforce decision in the firm (hiring, freezing, restructuring, outsourcing) rested on a single piece of arithmetic. It was so embedded in management thinking that nobody questioned it:
That equation no longer describes the world the firm operates in. AI did not improve it. AI broke it. The new equation looks like this:
This article covers what changed on each side of the equals sign, and what executives have to do this quarter to operate inside the new math.
Three questions, one equation
The new formula has three terms, and each one breaks something in traditional workforce planning:
- The left side: What are we measuring when we say “labor cost”?
- The multiplier: How is capability replacing headcount as the unit of value, and what does AI augmentation do to that math?
- The architecture: Why can’t current planning models operate this equation, and what infrastructure must replace them?
Let’s take them one at a time.
1. The left side: what we are measuring
In most organizations, the left-hand side of the equation (labor cost) is still treated as a static accounting figure. Salary, benefits, taxes, fully loaded cost per FTE. Clean. Structured. Completely insufficient.
Labor is not static. It moves, leaks, and compounds. Worse, it misallocates.
Labor often represents more than 50% of operating expenses. The bigger issue is opacity. Here is what typically stays invisible in the labor cost line:
- The cost of unfilled roles (vacancy drag)
- The cost of misaligned skills (productivity decay)
- The cost of turnover (replacement plus ramp inefficiency)
- The cost of poor allocation (right people, wrong work)
- The cost of not adopting AI in core processes
Recent industry analysis shows hidden costs from workforce misalignment could erase up to 15% of potential annual productivity gains in large organizations, and failure to embed AI in core functions carries opportunity costs of 5–12% of departmental operating budgets. These losses never reach the accounting record. They are strategic losses hidden in plain sight, on the wrong side of the equation.
Consider one specific number that has been visible in Latin American banking for years and that almost no payroll model captures. Between 2019 and 2025, headcount in retail banking grew approximately ten points faster than transactional volume. Ten points. That gap is a planning problem before it is a labor cost problem. Headcount kept compounding because the discipline that sized the workforce was looking at last year’s payroll line. It ignored the volume curves, the digital migration, the channel mix, and the capability gap opening between what the business needed and what the workforce could do.
The Latin American context further sharpens that opacity. In markets like Mexico, Colombia, or Brazil, severance and statutory costs can reach 30–40% of annual compensation. A poorly executed hiring freeze in these jurisdictions creates a deferred P&L event. The CFO discovers it eighteen months later, usually in a termination wave triggered by the next reorg. The “savings” are realizations of a liability that already existed. The number on the books moves. The economics stay where they were.
Or consider the inverse case. BGIS, a facilities management firm, protected margin without cutting headcount. It reduced voluntary attrition from 25% to 15% over two years by rebuilding how it measured what workforce cost meant. The left-hand side of their equation kept its size and became legible. That single shift (measurement before action) was worth more than a decade of hiring freezes.
That kind of move shows what happens when human capital stops being a line to minimize and starts being an asset to compound. The problem sits in the lens you use to read the left side of the equation.
2. The multiplier: from headcount to capability × (human + AI)
For decades, the right-hand side of the labor equation was a simple multiplication: cost per employee multiplied by the number of employees. AI replaces that multiplier outright.
The new multiplier has three properties that the old one did not.
First: capability replaces headcount as the unit of value. Two employees with identical salaries can now produce radically different outputs depending on the AI tools used, data accessibility, and process integration. The variance is no longer marginal. In client work across financial services, we have seen it exceed 40–60% within the same job family, grade, and compensation band. When you multiply cost by headcount, you treat both employees as identical units. When you multiply cost by capability, you finally see them as different economic objects.
Second: the parentheses matter more than the multiplier. (Human effort + AI augmentation) is a sum, and that sum is what produces output. In traditional planning, productivity gains were incremental and predictable. With AI augmentation, they are discontinuous. A single AI-assisted workflow redesign can eliminate entire task layers while increasing throughput. And yet 95% of enterprise AI pilots delivered zero measurable P&L impact. The organizations behind them failed to redesign work around the technology. The gains are real. The economics remain misunderstood.
Third: the equation has become a layered structure. Labor is no longer just labor. It is human cost, AI tooling cost, data infrastructure cost, and governance overhead. In most P&Ls, these sit in different categories. HR owns labor, IT owns tooling, FP&A owns the financial impact. Disconnected in planning, woven in reality. The equation has become multidimensional, while the chart of accounts remains flat.
The fix is organizational. Leadership teams that operate the new equation do three things consistently. They run joint planning sessions between HR, IT, and Finance; they set shared KPIs that measure workforce effectiveness and technology adoption together; and they treat workforce planning as a cross-functional discipline rather than an HR cycle. Bain’s research finds that companies taking this human-centric approach to productivity deliver more than 2x the total shareholder returns of those that do not.
The clearest sign that an organization is operating the new multiplier is whether its planning system can see capability as an object. It has to exist as a row, a dimension, a cell on a screen.
That radar chart is a working screen. The axes you see, the dimensions being tracked, are objects of the planning system. The difference between firms operating the new equation and firms still computing the old one shows up in whether their planning model has those axes.
The category error is assuming ROIC is calculated the same way before and after you lay off people to fund AI. The equation that justifies the cut is no longer the equation that evaluates the result.
3. The architecture: why FP&A cannot operate the new equation
Three weeks ago, in a bank in Miami, a CHRO opened a spreadsheet on the boardroom screen and said: “This is what HR sends to Finance every quarter.” Then she opened a second spreadsheet, from Finance back to her, and said: “This is what comes back.” The two files had different employee counts. They had been running on the same payroll for eleven years.
I have stopped being surprised by this. It happens almost every engagement.
The gap between those two spreadsheets looks like a data quality issue. It is the visible symptom of an older problem. Which discipline has the company installed to plan its workforce? And in 2026, that problem is what is producing the wave of AI-justified layoffs you have been reading about.
To see it, a distinction has to be made that most organizations have collapsed at their cost. Two disciplines tend to live under the same banner, “Planning,” yet they plan two different things.
FP&A (Financial Planning & Analysis) plans the financial statements. It plans the books. It lives in the general ledger, respects IFRS, closes by period, and forecasts what the P&L, balance sheet, and cash flow will look like under a set of assumptions. The question it answers is bounded and specific: “What will we report this quarter, and how do we close to it?” It is the discipline of planning the accounting record.
BP&A (Business Planning & Analytics) plans the business itself. It lives in capabilities, in skill inventories and capacity portfolios, and in the operational drivers that move the business (volume by channel, product mix, complexity, geography). The question it answers is messier: “How does the business actually produce value, and what combination of human capability, AI augmentation, and capital deployment compounds that value over time?” It is the discipline of planning operations and capacity.
I used to think this distinction was semantic. After more than thirty years in the field, I think it is one of the most consequential design choices a company makes. Most companies have made it without noticing.
Look at what the financial statements give you, and what they do not.
They give you the what. They tell you that labor cost rose 4.2%. That margins compressed 1.8 points. That the AI capex line ran where you projected it. That the headcount you were carrying produced a return that was, in financial reporting terms, exactly this. The financial statements are an after-image of the business. They are beautifully structured, tightly controlled, and designed for compliance and external reporting.
What they cannot tell you is the how. They cannot tell you which combination of people, skills, AI tools, and process design produced that 4.2% rise. They cannot tell you whether the people you let go in Q3 were the ones whose work an AI could replace, or whether they were the ones whose absence will tank a launch in Q2 of next year. They cannot tell you how to make human capital more productive when paired with technology that did not exist when your job architecture was designed. These are business questions, and the books were never asked to answer them.
In most organizations, FP&A has absorbed the territory of BP&A by default. No one else applied for the job. The accounting plan ate the business plan. Companies stopped planning the business and started planning the description of the business. The financial statements describe the result. The business plan determines whether there is a result worth describing. When the description starts to govern the underlying activity, you get what we have today. Workforce decisions look like cost discipline on the page. In operational reality, they slowly demolish capability.
That absorption is the structural cause of the current wave of AI-justified layoffs.
When revenue compresses or capex demands cash, FP&A does what FP&A is built to do. It cuts the most visible line on the statement it is planning. Headcount × salary is the calculation it owns. Headcount × salary is the lever it pulls. The decision is internally consistent. It is also structurally blind to almost everything that matters. FP&A has no way to evaluate the cost of not building capability, the delta in 24-month ROIC from a hiring delay, or the capability dependency between functions. None of those quantities exist on its surface. They are not in the chart of accounts. They cannot be reconciled to the GL. They have no cost center.
So FP&A produces the only diagnosis it knows how to produce, namely that we have too much labor cost on the statement we are about to report. That diagnosis is then dressed in strategic language (“AI transformation,” “operating model redesign,” “capability realignment”) and presented to the board as a considered choice.
Nothing about it was considered.
It is the mechanical output of statement-planning applied to a business-planning question, post-rationalized as strategy.
This is where the confrontation has to land, and where most CFOs will resist it. FP&A is excellent at what it is designed to do. The error is letting the discipline that plans the books also plan the business, and then accepting its output as if it described the business, when all it described was the books.
Decisions about capability (what to build, when to build it, what to retire, what to fuse with AI) are decisions about the productive structure of the firm. They are decisions about how human capital becomes more productive when strengthened by technological advances. They cannot be made on a discipline whose objects are line items and whose unit of analysis is the fiscal period. They require BP&A. They require planning the business itself instead of describing it after the fact.
The boards approving large workforce reductions today on the strength of FP&A models are doing something more uncomfortable than making the wrong decision. They are using statement-planning to make business-planning decisions, and reporting both inside the same discipline as if no error had occurred. The decisions go wrong because FP&A is being asked to plan something it was never designed to plan, by leadership teams who never installed the alternative.
The two planes that have to merge
I want to be specific about what the alternative looks like, because the abstraction has done enough work.
The architectural fix requires running two planes side by side, on the same surface, with their integrity preserved. Most organizations have one plane. They need two.
The strategic plane answers the capability term of the new equation. It asks what capabilities and skills the strategy requires by job family, geography, and time horizon, and what the risk is if those gaps stay open. It runs on operational drivers instead of payroll. In a bank, those drivers are volume by channel, product mix, transactional complexity, regulatory load. In a retailer, they are basket size, store format, omnichannel mix. In a manufacturer, they are throughput, SKU complexity, automation rate. The strategic plane is where business planning lives.
The financial-operational plane answers the cost × (human + AI) term. What does each capability decision cost, by entity, cost center, job family, and scenario? It is where BP&A and FP&A reconcile. Today the default is FP&A absorbing BP&A. Here, both are computable on the same surface, each with its integrity preserved. The business plan informs the statements. The statements do not invent the business plan.
Enterprise Performance Management platforms exist to articulate both planes without forcing one to swallow the other. Oracle EPM is the most complete example for this class of problem. Done right, the implementation lands in twelve to sixteen weeks. The technology matters, but the value is architectural. It gives BP&A its own tools, data layer, and scenarios, and only then reconciles them to the financial plane FP&A owns.
What you see above is a workspace. Each module is a discipline in its own right, with its own data, its own approvals, its own traceability. The interesting work happens in the conversations between modules. Here is one of them:
Look at the gesture. The difference between the two scenarios is fourteen million dollars on a single line, computed in seconds, defensible to the board. That is the minimum unit of BP&A planning. Producing the same view with FP&A would take three weeks, two consultants, and a footnote explaining the assumptions.
In our experience across Latin American banking and retail, this kind of architecture has produced combined impacts in the range of eight to fourteen percent of personnel expense, achieved without linear headcount cuts. It is the number you get when planning the business stops being a subroutine of planning the books.
Published case benchmarks are illustrative, with the caveat that they carry vendor and publication bias. KPMG reduced forecast days by 33%. Pearson collapsed budget revision cycles from three days to twelve hours. Securitas accelerated hiring speed by 70%. BGIS reduced voluntary attrition from 25% to 15% over two years. The pattern is consistent. When BP&A is given its own surface and stops being a subroutine inside FP&A, the planning cycle shortens, the data becomes defensible, and the conversation in the boardroom shifts from cost debates to capability decisions.
That shift, more than the technology, is what makes workforce planning strategic. It takes the willingness to stop letting the plan of the books govern decisions about how the business produces value.
Closing: three terms, three decisions
At the end of that boardroom conversation, the CFO did not ask for another report. He asked a harder question: “Can we model the real cost of how work actually gets done in this company?” He was, without using these words, asking to operate the new equation.
The discipline this requires can be summarized in three ideas, one per term.
- On the left side: labor cost. It is a composition, and you do not control what you cannot decompose. The ROIC equation has already moved capability into the asset column. Most charts of accounts have not.
- On the right side: the multiplier. Capability has replaced headcount as the unit of value. The parenthesis around (human + AI) carries the economic structure of every modern function.
- Around the equation: the architecture. EPM-class infrastructure is what makes the new formula computable. Without it, the equation stays on a slide.
This week, ask your team three questions, one per term:
- Left side: Is your labor cost line structured by capability (BP&A) or by cost center (FP&A)? Can you produce a single view of where your money is going by capability, today, without a special request to Finance?
- Multiplier: For your three most critical roles, what is the cost variance between top and median performers, and how much of that variance is explained by AI tool adoption? If you cannot answer this, you have a measurement problem.
- Architecture: Can HR, FP&A, and one business leader produce a single reconciled number for required vs. installed capacity in a single one-hour meeting?
If you cannot answer any of these cleanly, what you have is a payroll budget computing the wrong equation.
The 12 to 24 month horizon. Organizations that rebuild their workforce costing models on EPM-class architectures will operate on a different economic grammar than the rest. Published case benchmarks point to payback windows of 12–24 months and three-year ROI ranges of 80–250% (estimated, with meaningful variance by sector, data maturity, and implementation governance). The gap between leaders and laggards comes down to how quickly leadership teams accept two things. Workforce planning has moved from an HR cycle to a margin-protection discipline. And the equation that protects the margin is no longer the one they have been computing.
The companies that learn the new equation will go beyond better workforce planning. They will redefine what it means to generate ROIC once human capital stops being a cost in the denominator and starts being a capability that compounds the numerator.
The rest will keep multiplying headcount by salary.
And wondering, quietly, why the numbers don’t add up.
Pedro San Martín
Principal — Asher & Company
psanmartin@asher.company
